Molecular characterization of residual muscle-invasive bladder cancer identifies a scar-like transcriptomicprofile with favorable prognosis after neoadjuvant therapy

Abstract

Introduction: Many patients have residual disease after neoadjuvant therapy but their prognosis and need for adjuvant therapy is unclear. This study evaluates patient prognosis based on the molecular profiling of residual disease at cystectomy. Methods: RNA-seq data from TURBT samples was available from N=84 ABACUS patients of whom N=64 had matched RC samples, including 10 patients with pathologic complete response (pCR, ypT0N0). Pre- and post-atezolizumab tumor gene expression data was classified into molecular subtypes using the consensus subtyping model as a benchmark. Unsupervised consensus clustering (CC) was performed to categorize RC samples de novo, and each cluster was characterized using gene expression signatures. Two RC cohorts (NAC, N=133 & UTSW, N=94) known to harbor a scar-like biologic cluster were used for training and testing of a single-sample transcriptomic classifier (TC) that was validated in two independent (PURE-01, N=26 & ABACUS, N=64) RC cohorts after neoadjuvant immunotherapy. Results: Unsupervised CC revealed four distinct post-atezolizumab clusters (scar-like, basal, luminal-stromal & luminal). The scar-like cluster was present in 25% (16/64) of the post-atezolizumab samples and expressed genes associated with wound healing/scarring. A transcriptomic classifier (TC) trained to identify a favorable scar-like transcriptomic profile in residual bladder tumors showed robust performance in two validation cohorts, indicating a patient subgroup with favorable prognosis among NAC-treated, pembrolizumab-treated and atezolizumab-treated residual bladder cancer patients. Conclusions: This study contributes to the framework for defining molecular subtypes at radical cystectomy. Residual bladder cancer with a scar-like transcriptomic profile may predict favorable patient prognosis after neoadjuvant chemo- and immunotherapy.